Domain Ontology for Time Series Provenance

Abstract

Time series data are generated all the time with a volume without precedent, constituting themselves of a points sequence spread out over time, usually at time regular intervals. Time series analysis is different from data analysis, given its intrinsic nature, where observations are dependent and the observations order is important for analysis. The knowledge about the data which will be analyzed is relevant in an analysis process, but this knowledge is not always explicit and easy to interpret in many information resources. Time series can be semantically enriched where provenance information using ontologies allows to representing and inferring knowledge. The main contribution of this paper is to present a domain ontology developed by modular design for time series provenance, which adds semantic knowledge and contributes to the choice of appropriate statistical methods for an important step of time series analysis that is the trend extraction (detrending). Trend is a time series component that needs be extracted because it can hide other phenomena, as well as the most statistical methods are developed for stationary time series. With this work, is intended to contribute for semantically improving the decision making about trend extraction step, facilitating the preprocessing phase of time series analysis.